#503 – AI Crisis in Math: Navier-Stokes, Riemann Hypothesis & Complexity Theory

The mathematics community faces unprecedented disruption as artificial intelligence rapidly achieves feats once thought reserved for the most brilliant human minds, forcing researchers to reconsider the meaning of mathematical accomplishment and the very fabric of mathematical culture.

Ai Achievements in Mathematics: Scale and Speed

In recent weeks, AI systems have claimed solutions to some of the most profound longstanding mathematical challenges, including the Navier-Stokes equations, the Quasi-Riemann Hypothesis, and the Unique Games Conjecture. These advances are not mere incremental progress but represent dramatic leaps, compressing decades of human research into bursts of machine-generated discovery.

AI’s reach extends to the formal verification of mathematics as well. One of the most significant milestones is the auto-formalization and verification of Fermat’s Last Theorem in Lean—achieved in just 13 million lines of code with minimal human input. This accomplishment, anticipated to be years away, arrived swiftly: a product of AI systems building upon the structures of the MathLib library, directed by occasional human guidance but requiring little more than encouragement to “keep going.” The resulting proof is “sorry-free,” meaning that every logical step is verified without resorting to unproven assertions, and signals that auto-formalization is essentially a solved problem for existing mathematical results.

AI doesn’t just accelerate the solution process—it also alters expectations for future mathematical work. The sheer ability of an AI to gather, recombine, and apply existing mathematical knowledge at inhuman speeds leaves even top mathematicians—such as Terence Tao—both awed and alarmed. Tao expresses horror at how years of careful, patient research by graduate students and postdocs can be rendered obsolete instantly by an AI release, forever changing their experience of research and the landscape of open problems.

Distinguishing Between Solving Problems and Creating Understanding

AI systems demonstrate remarkable ability to blend and remix existing mathematical concepts in innovative ways, forming brilliant cascades of logical steps drawn from the current corpus of mathematical literature. However, mathematicians like Alex Kontorovich stress that while AI can reach astonishing breadth—connecting thousands of disparate ideas—and depth—resolving formidable problems—it fundamentally differs from human mathematicians in crucial ways.

First, AI approaches differ from traditional methods by often relying on statistical shortcuts rather than the kinds of intuitive, human-style insights and definitions that have historically driven true mathematical progress. For example, even the most celebrated AI solutions—the proof of the quasi-Riemann Hypothesis, for instance—are best viewed as extraordinary combinations of known mathematical moves rather than genuinely new ideas. The path taken by AI to solve such problems has been accessible within the mathematical “convex hull,” a metaphor implying that AI is synthesizing combinations of what humans already conceived, without shifting the very boundaries of the discipline.

However, this interpretation remains contested. Some, like Tim Gowers, prefer viewing AI progress as operating in a “semigroup” of ideas—a structure allowing for exponential growth by recombining known elements in ever-expanding ways. While AI may not yet generate fundamentally new mathematical definitions or concepts, its capacity to render the mathematics landscape transparent and accessible greatly outstrips human ability and paves the way for rapid further breakthroughs. Still, the distinction persists: only humans so far have pulled the “balloon” of mathematical knowledge outward by inventing entirely new concepts and expanding the terrain of what is possible.

AI’s methods, thus, can create a sense of accomplishment and progress, but risk bypassing the deeper human quest for “why”—the foundational understanding that gives mathematics its meaning. Mathematicians warn that, while AI can produce truths, understanding remains the domain of human creativity and reflection.

Threat to Traditional Mathematical Culture and Career Development

The arrival of AI as a solver and verifier of mathematical problems creates existential challenges for traditional modes of mathematical ...

Mathematics: Pursuit of Understanding Over Truth

Ankur Moitra and Alex Kontorovich emphasize that mathematics is far more than theorem-proving. Moitra notes that, compared to AI, human mathematicians are computationally limited and must rely on abstraction, conceptual frameworks, and generalizable insights to tackle major problems. This approach yields not only solutions but also systems that clarify connections between ideas, allowing mathematicians to “surround problems” and address families of related questions. Looking into mathematical papers, invisible threads emerge—networks of abstraction and understanding that extend beyond the proof of isolated statements.

For Moitra, the journey of fighting with a problem is as important as finding a solution. He cites Andrew Wiles as an example: the insight gained during the struggle is more cherished than any prize. Moitra criticizes the view of mathematics as “just a rising tide of theorems,” insisting that what truly matters is the system of beliefs and abstractions discovered in the process. As he illustrates with the story of von Neumann’s afterlife question—where the answer to the Riemann Hypothesis is less important than understanding why it is true—mathematicians crave the explanations and stories that proofs provide.

Both Moitra and Kontorovich stress the importance of mathematical storytelling. The process of understanding, learning why something is true, and translating this journey into narrative is central to transmitting knowledge. Kontorovich prefers to teach by intuition and calculation, introducing formal definitions only when necessary to solve a pressing problem. He predicts that future mathematical work will focus increasingly on the narrative around discoveries—what was learned, where the standard techniques failed or succeeded, and how new ideas emerged.

Telling the story of a proof’s development not only aids learning but also forms a bridge for others to tackle new problems. This narrative process, as Moitra says, is “more important than it ever has been.” Without it, the depth of discovery and value of human insight is lost, especially as technical problem-solving becomes more automatable.

Enduring Value of Human Insight in Problem Selection

Human mathematicians perform a crucial function that goes beyond solving technical problems: they select and frame problems worth investigating. Lex Fridman, Steven Brunton, and Alex Kontorovich discuss the unique human capability to direct attention to significant and deserving questions. Fridman describes this as a “shining a light” on the mathematical landscape. Brunton emphasizes that great mathematical work often begins by asking illuminating sub-questions, not just solving posed problems. Moitra notes that he often seeks the simplest unsolved problems, using them as entry points to develop tools and methods with broader applications.

The theorems and results that matter are those that the human mathematical community chooses to care about. Kontorovich observes that, almost tautologically, the theorems that become significant do so because humans find them meaningful, whether due to their deep connections, striking simplicity, or the beauty and elegance they reveal. Humans have the intuition to ask the right questions—sometimes guided by a mixture of beauty, difficulty, and curiosity—as in the great unsolved problems about prime numbers.

This ability to find simple, beautiful links, to create new definitions and forge connections across disparate areas, sets human mathematics apart. Kontorovich describes how introducing new concepts can expand the boundaries of mathematical knowledge in ways that mere recombination of existing results cannot.

In this framing, mathematics extends well beyond formal proofs: it includes forming theories, inventing concepts, assembling abstract frameworks, and revealing structural unity—a task that demands distinctly human judgment and insight.

Mathematics as Art and Expression of Beauty

Mathematics is celebrated not only for its utility but for its ...

AI’s role in mathematics is rapidly evolving, with mathematicians now considering whether AI should operate as a partner or a replacement. The "centaur world of math" reflects this new era, where mathematicians interact with intelligent systems, deciding which tasks to delegate to AI and what remains uniquely human work.

Centaur Mathematics as a Viable Future Path

The centaur model posits mathematics as a collaborative effort between humans and AI, rather than a solitary or solely mechanized pursuit.

Human-Ai Collaboration in Mathematics: Humans Direct, Ai Assists With Verification and Computation

Mathematicians like Ankur Moitra and Alex Kontorovich describe a future where humans guide mathematical inquiry and AI handles verification, computation, and exploring the technical landscape. In this collaborative paradigm, humans pose questions, direct investigations, and focus on the conceptual core, while AI supports by checking the validity of proofs and managing calculations.

Ai Excels At Exploring Possibilities, Identifying Patterns, and Suggesting Connections, Accelerating Understanding When Guided by Human Insight

AI can rapidly generate possibilities, identify patterns, and suggest unexpected connections across disparate fields or mathematical objects. When directed by human intuition, this accelerates the process of understanding and discovery, yet relies on human guidance to determine which connections and directions are worth pursuing.

Collaborative Approach vs. Chess: Preserving Human Roles in Math Through Community Agency In Questions and Theory

Unlike the fate of chess, where AI dominates, mathematics preserves significant human roles: defining worthwhile questions, devising methods, and building theories. The community’s collective agency in forming conjectures and prioritizing lines of investigation ensures humans remain essential to mathematical culture, even as AI becomes more capable.

The Solved Axiom and Its Contested Status

A central theoretical debate concerns the "Solved Axiom," which posits that any math problem solvable by a human-AI team could also be solved by AI alone.

Conjecture: Ai-alone Solvability of Human-Ai Problems Disputed, as Human Mathematical Practice Includes Non-automatable Elements

Alex Kontorovich does not subscribe fully to the Solved Axiom; he observes that although AI may mimic many mathematical tasks, much of human mathematical practice is exploratory and creative, particularly in crafting good questions and building deeper understanding. He recognizes AI’s growing capacity, even for asking and prioritizing questions, but stresses that uniquely human mathematical expertise and intuition still add distinctive value.

While It May Not Destroy Mathematics, It Would Shift Focus to Non-computational Aspects: Explaining Results, Building Theories, and Fostering Understanding

If AI ever realizes the Solved Axiom, mathematics would not be destroyed. Rather, the human focus would shift toward non-computational aspects such as explaining results, constructing coherent theoretical frameworks, and promoting genuine understanding—areas where formal computation alone cannot suffice.

Ai's Limits: Known Idea Combinations vs. New Concepts; Future Human Contribution in Math Uncertain

While AI can combine ideas from existing literature and suggest new directions, the creation of truly novel concepts or "theory-building" may resist automation. Kontorovich and Moitra both recognize the uncertainty in future human mathematical contributions as AI capabilities expand, but argue for an ongoing role rooted in human creativity, intuition, and explanatory skill.

Formalization and Collaboration Models for the Future

The integration of AI and collaborative tools is reshaping mathematical research, verification, and publication.

Lean Formalization and Platfor ...

The rapid integration of artificial intelligence into mathematics and research demands that educational and scholarly communities rethink traditional systems of validation, contribution, and training. Experts highlight the urgent need to adjust longstanding academic frameworks and refocus on human motivations as AI transforms every aspect of mathematical practice.

Reimagining how Mathematical Knowledge Is Validated and Shared

Journal System Strained, Needs Restructuring for Ai-generated Proofs and Deliberate Verification

Alex Kontorovich and Ankur Moitra note that mathematics journals and conferences are already overwhelmed by the sheer quantity of new papers, even before factoring in AI-generated research. Referees are exhausted, and the deliberate, precise process required to validate new results strains under volume. Now, journals face the additional task of reviewing AI-created proofs, but the community is hesitant to take on this work. There’s a clear need for new models—not to replace peer review, but to relieve the pressure and ensure mathematical epistemology remains careful and community-driven. Kontorovich emphasizes that while this slow process can be frustrating, it is still better than having no system for collective verification, recounting how the community took years to thoroughly check Perelman’s proof of the Poincaré conjecture, exemplifying the value of deep, collaborative deliberation.

Develop Alternative Mathematical Communication: Collaborative Platforms, Crowdsourced Proofs, and New Scholarly Recognition Beyond Traditional Metrics

Moitra supports developing collaborative efforts and crowdsourcing to evolve mathematical understanding and verification, contrasting it with the current system where draft ownership is individual and siloed. He envisions platforms that invite open participation and incremental improvement, with recognition for varied forms of scholarly contribution—such as code, datasets, reusable infrastructure, and APIs—which have become just as important as traditional papers, especially within large, multidisciplinary projects like LIGO or major cancer research efforts.

Engineering and Communicating Proofs: Ai vs. Human Expertise

Kontorovich distinguishes between engineering a proof (the formal, step-by-step construction) and communicating it effectively. He cites mathematical education, where intuition is often taught before formal construction, as a model for effective communication. AI may excel at generating or verifying formal arguments, but explaining the meaning and significance of a result in a way that fosters human understanding remains a deeply human challenge that must be valued and supported even as automation grows.

Transforming how We Value and Recognize Mathematical Contribution

Fields Medals Obsolete if Ai Solves Major Problems; Rethink Mathematics Celebrations

The panelists suggest that prizes like the Fields Medal, traditionally awarded for monumental breakthroughs, could become obsolete if AI begins to rapidly solve major problems. There is an open question about the future relevance of these awards, since their function has been to recognize insight, creativity, and perseverance—traits that may shift if algorithmic power dominates.

Recognize Value In Math: Beautiful Theories, Key Questions, Connections, and Future Mathematicians

Rather than simply celebrating the solving of longstanding conjectures, the community may evolve to value the identification of beautiful theories, posing of key questions, discovery of unexpected connections, and cultivation of future mathematicians. Steven Brunton emphasizes that mathematics should be about solving interesting problems and helping others, not just attaining fame or accumulating accolades.

Refocusing Success Metrics in Mathematics: Prioritizing Insight, Clarity, and Community Influence Over Publication Count and Speed

With the proliferation of AI and mass publication, the current incentives—such as publication counts and acceleration—are being questioned. The panelists stress the importance of refocusing metrics on depth of insight, clarity, and positive influence on the community, rather than on output volume or speed.

Rethinking Graduate Student Training and Career Paths

Traditional Mentoring Is Inadequate, Requiring Training in Questioning, Cross-Field Collaboration, and Intrinsic Motivation Development Independent of Achievement Markers

Moitra recounts a shift in his own mentorship, moving from assigning research questions within his field—which stifled student independence—to seeking areas slightly further afield, fostering student initiative and true intellectual ownership. He notes that the archetype of the researcher must change: training must go beyond technical expertise to encourage asking good questions, collaboration across fields, and cultivating intrinsic motivation—a drive to understand, not just to achieve.

Graduate Programs Must Prepare Students to Think Broadly and Synthesize Ideas Across Mathematics and Neighboring Disciplines

Moitra and others highlight that graduate programs should now intentionally train students to build bridges betwee ...

Navigating a future shaped by AI and rapid change prompts renewed investigation into what gives human life meaning, how we strive for fulfillment, and the role of learning and community in shaping a life worth living.

The Limitations of External Metrics For Human Flourishing

Steven Brunton observes that society has prized intelligence as the highest standard, conflating cognitive achievement with moral value. He cautions that superintelligence doesn't bring super morality, especially as highly intelligent AI systems may not act ethically. The current structure of education and professional success, he argues, perpetuates this misconception, pushing everyone toward college as a universal metric for value and insisting that rationalization and intelligence alone align with truth or justice.

Ankur Moitra highlights the "game to optimize" mindset, especially prominent in academic communities and competitive environments like Palo Alto. Students learn early to maximize their credentials—AP courses, the right extracurriculars, prestigious college admissions, and then professional interviews—continually chasing external validation. This cycle leaves many wondering why they're pursuing these milestones at all, as each achievement simply shifts the external goalposts. Moitra points out that while this external motivation may be sufficient in the beginning, accomplished individuals often find traditional success unsatisfying and must eventually search for an intrinsic reason to continue.

He suggests the future is unpredictable and that, rather than padding a CV, people fare much better by cultivating a sense of higher purpose and personal values. This approach enables adaptation and comfort with uncertainty as the world changes, guiding individuals toward their authentic blueprint for fulfillment.

Reclaiming the Intrinsic Value of the Learning Journey

Intrinsic motivation, curiosity, and the satisfaction of self-driven mastery are essential antidotes to the hollowness of external metrics. Kontorovich emphasizes how crucial it is to celebrate every bit of independence and self-sufficiency in learning: children glow with pride when mastering shoelaces—a seemingly small skill, but a major leap in autonomy. He argues that, even as AI can now deliver answers, students should prioritize the feeling of accomplishment from problem-solving over simply obtaining solutions.

Brunton and Kontorovich draw historical parallels from past technology adoption, such as calculators in mathematics. Although tools make certain skills redundant in practice, abandoning fundamentals has drawbacks—Kontorovich warns against skipping learning times tables just because a calculator can do the work, as this creates real deficiencies. Brunton echoes the importance of learning basics like linear algebra and calculus to nurture intuition and understanding, regardless of machines' prowess.

Education, Kontorovich argues, is most effective in the "zone of proximal development"—the space where a learner can progress with just enough challenge and support. AI can guide students in this space, checking understanding and motivating growth, but shouldn’t be a shortcut to bypass genuine learning. Fridman supports this, sharing that he finds joy and engagement in experimenting with concepts, learning through construction and play rather than rote answers.

Brunton underscores that the learning journey—the "road trip"—offers as much value as the destination itself. He encourages curiosity, deep learning, and resilience as the keys to satisfaction, urging educators to train students in critical advisor skills: navigating new information, questioning its source, and evaluating truth.

Recommitment to Human Connection and Community

The panelists agree that, especially as technology advances, the human need for connection cannot be substituted. Kontorovich predicts increased longing for human collaboration, mentorship, and interpersonal engagement, contrasting the recent decades’ push for technological autonomy. Fulfillment, he and Fridman argue, is ultimately grounded in meaningful relationships rather than career milestones.

Kontorovich shares personal perspective: while professional and academic achievements are gratifying, life’s deepest meaning is found through family and community. Fridman envisions that as AI commodifies intelligence, people will increasingly r ...